On solving trust-region and other regularised subproblems in optimization (Q977328): Difference between revisions

This paper revisits the popular Gay-Moré-Sorensen algorithm [\textit{D. M. Gay}; SIAM J. Sci. Stat. Comput. 2, 186--197 (1981; Zbl 0467.65027); \textit{J. J. Moré} and \textit{D. C. Sorensen}, SIAM J. Sci. Stat. Comput. 4, 553--572 (1983; Zbl 0551.65042)] for the direct solution of the trust-region subproblem using factorization in unconstrained optimization problems, and to provide flexible software for the related regularised quadratic subproblem. The optimality conditions for the trust-region subproblem is discussed, leading to a robust framework for its solution. Enhancements are provided such that the underlying method is both globally and superlinearly convergent in all cases. The ideas have been implemented as a pair of thread-safe Fortran 95 packages. Some numerical experiments are presented to show the effectiveness of the enhancements.